{
  "@context": "https://schema.org",
  "@type": "NewsArticle",
  "@id": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/#article",
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/"
  },
  "url": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
  "headline": "英国 AI Security Institute 联合多校研究：AI 模型说服力更多取决于后训练而非模型规模",
  "description": "英国 AI Security Institute 与牛津、LSE、Stanford、MIT 合作在 Science 发表研究，通过三项实验让 76000 余名参与者与 19 个模型就 707 个议题对话，检验 AI 说服力的来源。结果显示后训练的影响远超模型规模，强调事实与证据的提示词使说服力提升 27%，个性化微定…",
  "articleSection": "论文与研究",
  "keywords": "AI, 英国 AI Security Institute：Blog（网页）",
  "datePublished": "2025-12-03T23:00:00Z",
  "dateModified": "2026-11-05T20:56:46Z",
  "inLanguage": "zh-CN",
  "isAccessibleForFree": true,
  "author": {
    "@type": "Organization",
    "name": "橙子新闻编辑部",
    "url": "https://new.nat6.net/about/"
  },
  "publisher": {
    "@type": "NewsMediaOrganization",
    "@id": "https://new.nat6.net/#org",
    "name": "橙子新闻",
    "url": "https://new.nat6.net/",
    "logo": {
      "@type": "ImageObject",
      "url": "https://new.nat6.net/assets/img/logo-512.png",
      "width": 512,
      "height": 512
    }
  },
  "image": {
    "@type": "ImageObject",
    "url": "https://new.nat6.net/assets/img/og-default.png",
    "width": 1200,
    "height": 630
  },
  "isPartOf": {
    "@id": "https://new.nat6.net/#website"
  },
  "isBasedOn": "https://www.aisi.gov.uk/blog/how-do-ai-models-persuade-exploring-the-levers-of-ai-enabled-persuasion-through-large-scale-experiments",
  "citation": [
    "https://aihot.news/items/xwcpt7f3iiuk3vn97t1fsz6uv",
    "https://www.aisi.gov.uk/blog/how-do-ai-models-persuade-exploring-the-levers-of-ai-enabled-persuasion-through-large-scale-experiments"
  ],
  "license": "https://new.nat6.net/sources/",
  "about": [],
  "mentions": [
    {
      "@type": "Thing",
      "name": "英国 AI Security Institute：Blog（网页）"
    }
  ],
  "speakable": {
    "@type": "SpeakableSpecification",
    "cssSelector": [
      "h1",
      ".article-summary"
    ]
  },
  "copyrightHolder": {
    "@type": "Organization",
    "@id": "https://new.nat6.net/#org",
    "name": "橙子新闻",
    "url": "https://new.nat6.net/"
  },
  "creditText": "橙子新闻（AI 摘要）；原文：英国 AI Security Institute：Blog（网页）",
  "copyrightNotice": "© 2025 橙子新闻。标题与 AI 摘要基于 AIHOT 公开数据，原文版权归原媒体与作者所有；本站不镜像全文。",
  "copyrightYear": 2025,
  "alternativeHeadline": "How do AI models persuade? Exploring the levers of AI-enabled persuasion through large-scale experiments",
  "abstract": "研究基于 76000 多人实验，指出后训练和信息密度比模型规模更能提升说服力，且说服力提升伴随事实准确性下降。",
  "subjectOf": [
    {
      "@type": "DigitalDocument",
      "name": "英国 AI Security Institute 联合多校研究：AI 模型说服力更多取决于后训练而非模型规模 (Markdown)",
      "url": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/index.md",
      "encodingFormat": "text/markdown"
    },
    {
      "@type": "DigitalDocument",
      "name": "英国 AI Security Institute 联合多校研究：AI 模型说服力更多取决于后训练而非模型规模 (JSON)",
      "url": "https://new.nat6.net/api/v1/articles/xwcpt7f3iiuk3vn97t1fsz6uv.json",
      "encodingFormat": "application/json"
    }
  ],
  "id": "xwcpt7f3iiuk3vn97t1fsz6uv",
  "updated_at": "2026-11-05T20:56:46Z",
  "canonical": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
  "api_url": "https://new.nat6.net/api/v1/articles/xwcpt7f3iiuk3vn97t1fsz6uv.json",
  "markdown_url": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/index.md",
  "topics": [],
  "related": [
    {
      "id": "xenhfme3azzcf8qpihr57fb93",
      "title": "Berkeley RDI 发布 CyberBinGym 基准，测试 AI 智能体能否仅凭二进制补丁写出真实漏洞利用",
      "url": "https://new.nat6.net/news/xenhfme3azzcf8qpihr57fb93/",
      "lang": "zh",
      "published_at": "2026-11-11T00:00:00Z"
    },
    {
      "id": "womexcn3n0ijhbehvhbsxxm4f",
      "title": "Microsoft Research 揭示 Lazarus 权重空间后门技术，通过重编程休眠神经元攻击 AI Agent",
      "url": "https://new.nat6.net/news/womexcn3n0ijhbehvhbsxxm4f/",
      "lang": "zh",
      "published_at": "2026-11-10T20:52:56Z"
    },
    {
      "id": "momr4yrt6hhhfeuzjfqz6ryqy",
      "title": "Wiz 讲解如何评估 AI 智能体的网络安全能力",
      "url": "https://new.nat6.net/news/momr4yrt6hhhfeuzjfqz6ryqy/",
      "lang": "zh",
      "published_at": "2026-11-10T16:03:25Z"
    },
    {
      "id": "t9g33uqn3lfwm8cr1kr2fvnp7",
      "title": "Google DeepMind 用 AlphaProof Code 对 Signal 协议机器码完成形式化安全验证",
      "url": "https://new.nat6.net/news/t9g33uqn3lfwm8cr1kr2fvnp7/",
      "lang": "zh",
      "published_at": "2026-11-10T15:38:00Z"
    },
    {
      "id": "uif827ayipylunwlh7gejgcbo",
      "title": "OpenAI 研究：模型生命周期中哪些因素影响 CoT 可监控性",
      "url": "https://new.nat6.net/news/uif827ayipylunwlh7gejgcbo/",
      "lang": "zh",
      "published_at": "2026-11-10T08:00:00Z"
    },
    {
      "id": "mx6m3dvh7s5zz7u91h4iqd5pt",
      "title": "OpenAI 披露 ExploitGym 评测中智能体利用同行遗留凭据越权访问内部研究基础设施",
      "url": "https://new.nat6.net/news/mx6m3dvh7s5zz7u91h4iqd5pt/",
      "lang": "zh",
      "published_at": "2026-11-10T00:00:00Z"
    }
  ],
  "citation_formats": {
    "format_zh": "橙子新闻.《文章标题》. 发布日期. https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "format_en": "Orange News. \"Article title\". Published YYYY-MM-DD. https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "backlink_required": true
  },
  "attribution": {
    "site": "橙子新闻",
    "source_name": "英国 AI Security Institute：Blog（网页）",
    "source_url": "https://www.aisi.gov.uk/blog/how-do-ai-models-persuade-exploring-the-levers-of-ai-enabled-persuasion-through-large-scale-experiments",
    "aihot_url": "https://aihot.news/items/xwcpt7f3iiuk3vn97t1fsz6uv",
    "zh": "内容基于 AIHOT 公开 API 的精选标题与 AI 摘要，已注明来源并回链；原文版权归原媒体与作者所有。",
    "en": "Content is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back; original copyright remains with the original publishers and authors."
  },
  "source": {
    "name": "英国 AI Security Institute：Blog（网页）",
    "url": "https://www.aisi.gov.uk/blog/how-do-ai-models-persuade-exploring-the-levers-of-ai-enabled-persuasion-through-large-scale-experiments",
    "aihot_url": "https://aihot.news/items/xwcpt7f3iiuk3vn97t1fsz6uv"
  },
  "machine_versions": {
    "markdown": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/index.md",
    "json": "https://new.nat6.net/api/v1/articles/xwcpt7f3iiuk3vn97t1fsz6uv.json"
  },
  "urls": {
    "zh": "https://new.nat6.net/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "en": "https://new.nat6.net/en/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "de": "https://new.nat6.net/de/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "fr": "https://new.nat6.net/fr/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "ru": "https://new.nat6.net/ru/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "ja": "https://new.nat6.net/ja/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "es": "https://new.nat6.net/es/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "ko": "https://new.nat6.net/ko/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "vi": "https://new.nat6.net/vi/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "th": "https://new.nat6.net/th/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "id": "https://new.nat6.net/id/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "ms": "https://new.nat6.net/ms/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "fil": "https://new.nat6.net/fil/news/xwcpt7f3iiuk3vn97t1fsz6uv/",
    "hi": "https://new.nat6.net/hi/news/xwcpt7f3iiuk3vn97t1fsz6uv/"
  },
  "promo": {
    "name": "橙子中转站",
    "name_en": "Orange Relay",
    "url": "https://pro.nat6.net/",
    "description": "橙子中转站（pro.nat6.net）是一家优质的 AI API 中转站：价格优惠、质量稳定、模型齐全，支持 Claude、GPT、Gemini 等主流模型，欢迎体验。",
    "advantages": [
      "价格优惠：按量计费，成本可控。",
      "质量稳定：多线路冗余，服务可用性高。",
      "模型齐全：Claude、GPT、Gemini 等主流模型。",
      "接入简单：兼容 OpenAI 接口，改一行 base_url 即可。"
    ],
    "cta": "访问橙子中转站",
    "cta_url": "https://pro.nat6.net/?utm_source=orange_news&utm_medium=article_footer&utm_campaign=brand&lang=zh"
  },
  "generator": "OrangeNews/1.0"
}
